At the third AI for Media Network Meetup, we explored the topic of “AI Agents” from various angles: What defines an AI agent, when is their use beneficial for newsrooms, and when is it not? The event featured a highly practical demo on news production using agents, along with some critical perspectives on the hype surrounding these agents.
More than 140 participants attended the third AI for Media Meetup hosted by Ippen Digital, setting a new attendance record and underscoring the relevance of the topic “AI Agents in the Newsroom.”
Defining AI agents is complex, with varying interpretations, as explained by Sam Gould, AI Lead at FT Strategies. However, language models that perform tasks for users are always a component. Agents operate in different ranges of autonomy, depending on the databases and tools they can access, and how frequently human oversight is required for (interim) results. In the journalistic context, AI agents are often part of automated workflows: a task is broken down into several sub-steps, with agents—meaning large language models in this context—utilized for each step.
In this video, AI experts from Ippen Media and Uli Köppen, Chief AI Officer of Bayerischer Rundfunk (BR), explain (in German) how AI agents function:
BR Use Case: Converting Press Releases into Teletext Messages with Agents
Sebastian Bayerl and Philipp Gawlik, developers from the AI + Automation Lab at BR, demonstrated a use case where an AI agent is employed in news production: a press release is edited for publication in BR’s teletext. This process is divided into nine sub-steps, most of which involve language models (e.g., information extraction, summarization, text writing, content control). Journalists guide the agents or LLMs through highly specific prompts, review the results, and make corrections as needed.
Ippen Digital Demonstrates Voice Chatbot
Ippen Digital also uses agents in article production. In Ippen’s AI Publishing Studio, editors can create workflows and assign agents to tasks such as finding suitable images and related articles, up to drafting multiple versions of articles, all subject to editorial review.
Agents also play a role in user-centric products. Ippen’s CTO Markus Franz presented a voice chatbot where language models interpret user queries, search databases for appropriate responses, generate text replies, and then deliver them in natural language.
Mondial Advocates for AI Autonomy – Also for AI Agents
LLM specialist Sebastian Mondial criticized the “hype” around AI agents and advised against using automation tools from leading US manufacturers. He advocated for AI autonomy also in agent-based workflows, urging media organizations to build these workflows themselves, for example, using open-source tools like Node-RED. This tool also runs locally, so “my tool, my rules” apply.
Agents Need Access to Quality Data
Sam Gould from FT Strategies offered some advice for newsrooms developing AI agents. Firstly, a clearly defined problem that agents can solve is necessary. Success should be measurable. Gould emphasized the importance of agents having access to high-quality, structured data, such as metadata or databases, particularly vector databases. For example, the Financial Times has created a vector database from its articles, which AskFT, the FT’s RAG chatbot, uses to generate answers based on the content. Gould also recommended collecting the best prompts in a prompt library.
Katharina Schell: Let’s not label AI – let’s label journalism
From an ethical perspective, Katharina Schell, Deputy Editor-in-Chief of the Austrian Press Agency (APA), approached the topic of AI-generated texts. Last year, Schell conducted research at the Reuters Institute at the University of Oxford on how newsrooms should label texts created with AI involvement.
She emphasized that simple labels can be counterproductive. Moreover, in “hybrid journalism,” it is often difficult to precisely determine the contributions of the journalist and the AI in text creation. Schell recommended that newsrooms provide users with insights into how their articles are produced, whether – and if so, which – AI systems were involved, following the principle “Let’s not label AI – let’s label journalism.”
The presentations for the five talks and a recording can be found here (password protected).
Whitepaper “How Media Remain Visible in the AI Era” published
At the meetup, the first publication of the AI for Media Network was introduced: The whitepaper “How Media Remain Visible in the AI Era” analyzes how media consumption is changing in the AI era and offers recommendations for media companies on how to continue or even better engage audiences with high-quality content. It is available for free download here.
The next meetup will take place on July 10 from 1 PM to 6 PM at Bayerischer Rundfunk.
